RDoC-informed explainable AI as a paradigm for multilevel Alzheimer's disease diagnosis and progression prediction: a...

Area of Science:

• Neuroscience
• Artificial Intelligence
• Biomarker Discovery

Background:

• Explainable Artificial Intelligence (XAI) is increasingly utilized for dementia diagnosis and monitoring.
• The National Institute of Mental Health's Research Domain Criteria (NIMH-RDoC) framework offers a dimensional structure for analyzing mental disorders across various biological and behavioral levels.

Purpose of the Study:

• To propose integrating the NIMH-RDoC framework with XAI-informed diagnostic protocols for early Alzheimer's disease (AD) diagnosis.
• To critically analyze current XAI applications in dementia research, focusing on diagnostic and prognostic capabilities.
• To enhance the mechanistic validity and interpretability of data-driven models in AD research.

Main Methods:

• Restructuring diverse input features (genetics, biomarkers, neuroimaging, electrophysiology, cognition, behavior) onto RDoC units.
• Applying XAI techniques to analyze and interpret complex datasets for AD diagnosis.
• Grounding XAI explanations within RDoC cognitive domains for neuropsychological relevance.

Main Results:

• The proposed 'converging RDoC and XAI' approach provides more insightful and inclusive diagnostic models.
• This integration enhances the mechanistic validity of data-driven models by aligning them with neurobiological domains.
• The framework facilitates a strategic roadmap for translational neuroscience and personalized medicine in AD.

Conclusions:

• Combining RDoC and XAI bolsters AD biomarker coherence and aids clinical decision-making.
• This integrated approach supports neuropsychologically-informed diagnosis and communication for patients and caregivers.
• The framework offers a novel strategy for advancing early AD detection and personalized treatment.